authorgravatar for jacobly@ziglang.orgJacob Young <jacobly@ziglang.org> 2023-05-02 03:24:04-04:00
committergravatar for jacobly@ziglang.orgJacob Young <jacobly@ziglang.org> 2023-05-03 04:25:14-04:00
log9ccdbca635a3b5a26b65ab8e52533d3acc8f2f5e
treeda608c7ad4fc308a7707863d7cc917e4af41b02f
parent31429a4e8649961624878d11e1bb330107013086

x86_64: implement fabs

6 files changed, 52 insertions(+), 7 deletions(-)

src/arch/x86_64/CodeGen.zig+14-6
...@@ -1458,14 +1458,13 @@ fn genBody(self: *Self, body: []const Air.Inst.Index) InnerError!void {...@@ -1458,14 +1458,13 @@ fn genBody(self: *Self, body: []const Air.Inst.Index) InnerError!void {
1458 .log,1458 .log,
1459 .log2,1459 .log2,
1460 .log10,1460 .log10,
1461 .fabs,
1462 .floor,1461 .floor,
1463 .ceil,1462 .ceil,
1464 .round,1463 .round,
1465 .trunc_float,1464 .trunc_float,
1466 => try self.airUnaryMath(inst),1465 => try self.airUnaryMath(inst),
14671466
1468 .neg => try self.airNeg(inst),1467 .neg, .fabs => try self.airFloatSign(inst),
14691468
1470 .add_with_overflow => try self.airAddSubWithOverflow(inst),1469 .add_with_overflow => try self.airAddSubWithOverflow(inst),
1471 .sub_with_overflow => try self.airAddSubWithOverflow(inst),1470 .sub_with_overflow => try self.airAddSubWithOverflow(inst),
...@@ -4185,7 +4184,7 @@ fn airBitReverse(self: *Self, inst: Air.Inst.Index) !void {...@@ -4185,7 +4184,7 @@ fn airBitReverse(self: *Self, inst: Air.Inst.Index) !void {
4185 return self.finishAir(inst, dst_mcv, .{ ty_op.operand, .none, .none });4184 return self.finishAir(inst, dst_mcv, .{ ty_op.operand, .none, .none });
4186}4185}
41874186
4188fn airNeg(self: *Self, inst: Air.Inst.Index) !void {4187fn airFloatSign(self: *Self, inst: Air.Inst.Index) !void {
4189 const un_op = self.air.instructions.items(.data)[inst].un_op;4188 const un_op = self.air.instructions.items(.data)[inst].un_op;
4190 const ty = self.air.typeOf(un_op);4189 const ty = self.air.typeOf(un_op);
4191 const ty_bits = ty.floatBits(self.target.*);4190 const ty_bits = ty.floatBits(self.target.*);
...@@ -4228,10 +4227,19 @@ fn airNeg(self: *Self, inst: Air.Inst.Index) !void {...@@ -4228,10 +4227,19 @@ fn airNeg(self: *Self, inst: Air.Inst.Index) !void {
4228 const dst_lock = self.register_manager.lockReg(dst_mcv.register);4227 const dst_lock = self.register_manager.lockReg(dst_mcv.register);
4229 defer if (dst_lock) |lock| self.register_manager.unlockReg(lock);4228 defer if (dst_lock) |lock| self.register_manager.unlockReg(lock);
42304229
4230 const tag = self.air.instructions.items(.tag)[inst];
4231 try self.genBinOpMir(switch (ty_bits) {4231 try self.genBinOpMir(switch (ty_bits) {
4232 32 => .xorps,4232 32 => switch (tag) {
4233 64 => .xorpd,4233 .neg => .xorps,
4234 else => return self.fail("TODO implement airNeg for {}", .{4234 .fabs => .andnps,
4235 else => unreachable,
4236 },
4237 64 => switch (tag) {
4238 .neg => .xorpd,
4239 .fabs => .andnpd,
4240 else => unreachable,
4241 },
4242 else => return self.fail("TODO implement airFloatSign for {}", .{
4235 ty.fmt(self.bin_file.options.module.?),4243 ty.fmt(self.bin_file.options.module.?),
4236 }),4244 }),
4237 }, vec_ty, dst_mcv, sign_mcv);4245 }, vec_ty, dst_mcv, sign_mcv);
src/arch/x86_64/Encoding.zig+6
...@@ -268,23 +268,29 @@ pub const Mnemonic = enum {...@@ -268,23 +268,29 @@ pub const Mnemonic = enum {
268 movd,268 movd,
269 // SSE269 // SSE
270 addss,270 addss,
271 andps,
272 andnps,
271 cmpss,273 cmpss,
272 cvtsi2ss,274 cvtsi2ss,
273 divss,275 divss,
274 maxss, minss,276 maxss, minss,
275 movss,277 movss,
276 mulss,278 mulss,
279 orps,
277 subss,280 subss,
278 ucomiss,281 ucomiss,
279 xorps,282 xorps,
280 // SSE2283 // SSE2
281 addsd,284 addsd,
285 andpd,
286 andnpd,
282 //cmpsd,287 //cmpsd,
283 cvtsd2ss, cvtsi2sd, cvtss2sd,288 cvtsd2ss, cvtsi2sd, cvtss2sd,
284 divsd,289 divsd,
285 maxsd, minsd,290 maxsd, minsd,
286 movq, //movd, movsd,291 movq, //movd, movsd,
287 mulsd,292 mulsd,
293 orpd,
288 subsd,294 subsd,
289 ucomisd,295 ucomisd,
290 xorpd,296 xorpd,
src/arch/x86_64/Lower.zig+6
...@@ -94,6 +94,8 @@ pub fn lowerMir(lower: *Lower, inst: Mir.Inst) Error![]const Instruction {...@@ -94,6 +94,8 @@ pub fn lowerMir(lower: *Lower, inst: Mir.Inst) Error![]const Instruction {
94 .xor,94 .xor,
9595
96 .addss,96 .addss,
97 .andnps,
98 .andps,
97 .cmpss,99 .cmpss,
98 .cvtsi2ss,100 .cvtsi2ss,
99 .divss,101 .divss,
...@@ -101,11 +103,14 @@ pub fn lowerMir(lower: *Lower, inst: Mir.Inst) Error![]const Instruction {...@@ -101,11 +103,14 @@ pub fn lowerMir(lower: *Lower, inst: Mir.Inst) Error![]const Instruction {
101 .minss,103 .minss,
102 .movss,104 .movss,
103 .mulss,105 .mulss,
106 .orps,
104 .roundss,107 .roundss,
105 .subss,108 .subss,
106 .ucomiss,109 .ucomiss,
107 .xorps,110 .xorps,
108 .addsd,111 .addsd,
112 .andnpd,
113 .andpd,
109 .cmpsd,114 .cmpsd,
110 .cvtsd2ss,115 .cvtsd2ss,
111 .cvtsi2sd,116 .cvtsi2sd,
...@@ -115,6 +120,7 @@ pub fn lowerMir(lower: *Lower, inst: Mir.Inst) Error![]const Instruction {...@@ -115,6 +120,7 @@ pub fn lowerMir(lower: *Lower, inst: Mir.Inst) Error![]const Instruction {
115 .minsd,120 .minsd,
116 .movsd,121 .movsd,
117 .mulsd,122 .mulsd,
123 .orpd,
118 .roundsd,124 .roundsd,
119 .subsd,125 .subsd,
120 .ucomisd,126 .ucomisd,
src/arch/x86_64/Mir.zig+12
...@@ -168,6 +168,10 @@ pub const Inst = struct {...@@ -168,6 +168,10 @@ pub const Inst = struct {
168168
169 /// Add single precision floating point values169 /// Add single precision floating point values
170 addss,170 addss,
171 /// Bitwise logical and of packed single precision floating-point values
172 andps,
173 /// Bitwise logical and not of packed single precision floating-point values
174 andnps,
171 /// Compare scalar single-precision floating-point values175 /// Compare scalar single-precision floating-point values
172 cmpss,176 cmpss,
173 /// Convert doubleword integer to scalar single-precision floating-point value177 /// Convert doubleword integer to scalar single-precision floating-point value
...@@ -182,6 +186,8 @@ pub const Inst = struct {...@@ -182,6 +186,8 @@ pub const Inst = struct {
182 movss,186 movss,
183 /// Multiply scalar single-precision floating-point values187 /// Multiply scalar single-precision floating-point values
184 mulss,188 mulss,
189 /// Bitwise logical or of packed single precision floating-point values
190 orps,
185 /// Round scalar single-precision floating-point values191 /// Round scalar single-precision floating-point values
186 roundss,192 roundss,
187 /// Subtract scalar single-precision floating-point values193 /// Subtract scalar single-precision floating-point values
...@@ -192,6 +198,10 @@ pub const Inst = struct {...@@ -192,6 +198,10 @@ pub const Inst = struct {
192 xorps,198 xorps,
193 /// Add double precision floating point values199 /// Add double precision floating point values
194 addsd,200 addsd,
201 /// Bitwise logical and not of packed double precision floating-point values
202 andnpd,
203 /// Bitwise logical and of packed double precision floating-point values
204 andpd,
195 /// Compare scalar double-precision floating-point values205 /// Compare scalar double-precision floating-point values
196 cmpsd,206 cmpsd,
197 /// Convert scalar double-precision floating-point value to scalar single-precision floating-point value207 /// Convert scalar double-precision floating-point value to scalar single-precision floating-point value
...@@ -210,6 +220,8 @@ pub const Inst = struct {...@@ -210,6 +220,8 @@ pub const Inst = struct {
210 movsd,220 movsd,
211 /// Multiply scalar double-precision floating-point values221 /// Multiply scalar double-precision floating-point values
212 mulsd,222 mulsd,
223 /// Bitwise logical or of packed double precision floating-point values
224 orpd,
213 /// Round scalar double-precision floating-point values225 /// Round scalar double-precision floating-point values
214 roundsd,226 roundsd,
215 /// Subtract scalar double-precision floating-point values227 /// Subtract scalar double-precision floating-point values
src/arch/x86_64/encodings.zig+12
...@@ -832,6 +832,10 @@ pub const table = [_]Entry{...@@ -832,6 +832,10 @@ pub const table = [_]Entry{
832 // SSE832 // SSE
833 .{ .addss, .rm, &.{ .xmm, .xmm_m32 }, &.{ 0xf3, 0x0f, 0x58 }, 0, .sse },833 .{ .addss, .rm, &.{ .xmm, .xmm_m32 }, &.{ 0xf3, 0x0f, 0x58 }, 0, .sse },
834834
835 .{ .andnps, .rm, &.{ .xmm, .xmm_m128 }, &.{ 0x0f, 0x55 }, 0, .sse },
836
837 .{ .andps, .rm, &.{ .xmm, .xmm_m128 }, &.{ 0x0f, 0x54 }, 0, .sse },
838
835 .{ .cmpss, .rmi, &.{ .xmm, .xmm_m32, .imm8 }, &.{ 0xf3, 0x0f, 0xc2 }, 0, .sse },839 .{ .cmpss, .rmi, &.{ .xmm, .xmm_m32, .imm8 }, &.{ 0xf3, 0x0f, 0xc2 }, 0, .sse },
836840
837 .{ .cvtsi2ss, .rm, &.{ .xmm, .rm32 }, &.{ 0xf3, 0x0f, 0x2a }, 0, .sse },841 .{ .cvtsi2ss, .rm, &.{ .xmm, .rm32 }, &.{ 0xf3, 0x0f, 0x2a }, 0, .sse },
...@@ -848,6 +852,8 @@ pub const table = [_]Entry{...@@ -848,6 +852,8 @@ pub const table = [_]Entry{
848852
849 .{ .mulss, .rm, &.{ .xmm, .xmm_m32 }, &.{ 0xf3, 0x0f, 0x59 }, 0, .sse },853 .{ .mulss, .rm, &.{ .xmm, .xmm_m32 }, &.{ 0xf3, 0x0f, 0x59 }, 0, .sse },
850854
855 .{ .orps, .rm, &.{ .xmm, .xmm_m128 }, &.{ 0x0f, 0x56 }, 0, .sse },
856
851 .{ .subss, .rm, &.{ .xmm, .xmm_m32 }, &.{ 0xf3, 0x0f, 0x5c }, 0, .sse },857 .{ .subss, .rm, &.{ .xmm, .xmm_m32 }, &.{ 0xf3, 0x0f, 0x5c }, 0, .sse },
852858
853 .{ .ucomiss, .rm, &.{ .xmm, .xmm_m32 }, &.{ 0x0f, 0x2e }, 0, .sse },859 .{ .ucomiss, .rm, &.{ .xmm, .xmm_m32 }, &.{ 0x0f, 0x2e }, 0, .sse },
...@@ -857,6 +863,10 @@ pub const table = [_]Entry{...@@ -857,6 +863,10 @@ pub const table = [_]Entry{
857 // SSE2863 // SSE2
858 .{ .addsd, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x58 }, 0, .sse2 },864 .{ .addsd, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x58 }, 0, .sse2 },
859865
866 .{ .andnpd, .rm, &.{ .xmm, .xmm_m128 }, &.{ 0x66, 0x0f, 0x55 }, 0, .sse2 },
867
868 .{ .andpd, .rm, &.{ .xmm, .xmm_m128 }, &.{ 0x66, 0x0f, 0x54 }, 0, .sse2 },
869
860 .{ .cmpsd, .rmi, &.{ .xmm, .xmm_m64, .imm8 }, &.{ 0xf2, 0x0f, 0xc2 }, 0, .sse2 },870 .{ .cmpsd, .rmi, &.{ .xmm, .xmm_m64, .imm8 }, &.{ 0xf2, 0x0f, 0xc2 }, 0, .sse2 },
861871
862 .{ .cvtsd2ss, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x5a }, 0, .sse2 },872 .{ .cvtsd2ss, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x5a }, 0, .sse2 },
...@@ -883,6 +893,8 @@ pub const table = [_]Entry{...@@ -883,6 +893,8 @@ pub const table = [_]Entry{
883893
884 .{ .mulsd, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x59 }, 0, .sse2 },894 .{ .mulsd, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x59 }, 0, .sse2 },
885895
896 .{ .orpd, .rm, &.{ .xmm, .xmm_m128 }, &.{ 0x66, 0x0f, 0x56 }, 0, .sse2 },
897
886 .{ .subsd, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x5c }, 0, .sse2 },898 .{ .subsd, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x5c }, 0, .sse2 },
887899
888 .{ .movsd, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x10 }, 0, .sse2 },900 .{ .movsd, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x10 }, 0, .sse2 },
test/behavior/floatop.zig+2-1
...@@ -96,7 +96,8 @@ test "negative f128 floatToInt at compile-time" {...@@ -96,7 +96,8 @@ test "negative f128 floatToInt at compile-time" {
96}96}
9797
98test "@sqrt" {98test "@sqrt" {
99 if (builtin.zig_backend == .stage2_x86_64) return error.SkipZigTest; // TODO99 if (builtin.zig_backend == .stage2_x86_64 and
100 comptime !std.Target.x86.featureSetHasAll(builtin.cpu.features, .{ .sse, .sse2, .sse4_1 })) return error.SkipZigTest; // TODO
100 if (builtin.zig_backend == .stage2_arm) return error.SkipZigTest; // TODO101 if (builtin.zig_backend == .stage2_arm) return error.SkipZigTest; // TODO
101 if (builtin.zig_backend == .stage2_aarch64) return error.SkipZigTest; // TODO102 if (builtin.zig_backend == .stage2_aarch64) return error.SkipZigTest; // TODO
102 if (builtin.zig_backend == .stage2_sparc64) return error.SkipZigTest; // TODO103 if (builtin.zig_backend == .stage2_sparc64) return error.SkipZigTest; // TODO